Arbaz Boltx\
Agentic AI Engineer
Talent
LahoreMember since October 14, 2026
LoRAQLoRAPEFTQwenLlamaDeepSeek4/8-bit quantizationCUDA optimizationLangChainLangGraphLangSmithLangfuseMCPtool-callingRAGn8nZapierDeepgramCartesiaWhisperDjangoDRFJWTOAuthwebhooksNext.js 15RunPodHetznerinference serving
·
Hire this person
Send a job offer directly to this candidate
LoRAQLoRAPEFTQwenLlama
About
Agentic AI engineer with 4 years of experience building autonomous agent loops, fine-tuning open-weight LLMs (Qwen, Llama, DeepSeek) with LoRA/QLoRA/PEFT and 4/8-bit quantization, and serving them on self-hosted GPUs. Ships low-latency voice agents, tool-calling and RAG systems (LangGraph, MCP) and multi-tenant SaaS on Django/DRF.
Experience
AI Instructor (Part-time) at PNY Trainings (2026-09 – Present)
- Teach weekend courses on n8n, Zapier, LangChain, LangGraph, LangSmith and Langfuse, with a focus on foundational AI understanding.
AI Engineer at IPS USA (IPS BPO) (2026-06 – Present)
- Built a MIPS RAG system for medical billing: fine-tuned DeepSeek R1 with PEFT, ran it locally, and exposed a JSON-in, structured-JSON-out script.
- Built a QA call evaluation portal: Deepgram speech-to-text with an NVIDIA Parakeet 0.6B fallback and a local Qwen 7B model as the LLM.
- Fine-tuned a model with PEFT for Truckverse and am deploying it end to end on GPUs via RunPod.
- Building a multi-tenant CRM SaaS (Django/DRF, Next.js 15, LangGraph, PostgreSQL tenant isolation) and an agentic plan take-off system that generates Excel take-offs from drawings.
Senior Voice Engineer (Freelance) at Goldman World Logistics (2026-05 – 2026-08)
- Evaluated open-source voice stacks (Dograh, Whisper, Cartesia SDK, NVIDIA Nemotron, Kyutai Unmute) and patched voice agents for concurrency.
- Built a car-transport dispatch voice agent that calls customers, pulls quotes from the Super Dispatch API, books on the call, and transfers to a human agent.
AI Engineer (Freelance) at Turing (2026-01 – 2026-07)
- Trained Cursor's Composer model for frontier AI labs, including fine-tuning Composer 2.5 for domain-specific tasks.
- Built image and coding evaluation pipelines benchmarked against human-graded rubrics to catch regressions before release.
- Applied CUDA optimization and 4-bit quantization to cut GPU inference cost across coding and voice agents.
Django Developer (Contract) at Devsecmlops (2024-02 – 2025-09)
- Built Letshab, a voice agent platform: Django/DRF backend with tool-calling so live calls trigger database writes, calendar bookings and CRM updates.
- Owned the multi-tenant auth system (JWT) and integrated 6 CRM systems via webhooks and OAuth.
- Handled model training, quantization and inference serving; tuned concurrency and throughput for production call volumes.
Education
BSc in Business (2026)
Associate in Computer Science – APS
Skills
Agentic AI Engineer

